Research On SIFT - Based Panoramic Image Mosaic Method

Panoramic image stitching is proposed to solve the problem that the ordinary imaging devices is difficult to captured high-resolution wide-field image’s due to the limited perspective, it has become a research focus in the area of computer vision, digital image processing etc. With the development of computer technology, image stitching technology is widely used in aerospace image processing, medical image processing, ultra-wide image mosaic, image mosaic pieces, virtual reality and video compression.In this paper, image which is captured by hand-held camera with the feature of translation, rotation, scaling, perspective transformation and complex changes is used as study, and the theory and methods of panoramic image stitching based on feature is explored. Key perspective in splicing processes such as image registration, image conversion and image fusion is also illustrated. Next, this paper analyses and summarizes the main issues of principles, methods, performance as well as the existed problems.Among of them, image registration phase is studied based on the classic SIFT algorithm. This paper designs a Three times SIFT registration method for image sequence by means of the good robustness of classic SIFT algorithm to with rotation and scale invariance, perspective transformation, complex situation as under noise and light changes. This method can not only reduce the matching time greatly but also ensure correct match rate at sometime. In image fusion stage, weighted average fusion method which based on a comprehensive comparison of performance and time spent is selected. Finally, image is projected onto a cylindrical panoramic model, further to complete the whole process based on cylindrical panoramic image stitching. Experimental results show that the improved algorithm can not only shorten the stitching time, but also acquire a better image quality. Meanwhile, it has has a high correct matching rate.